Project

General

Profile

Actions

Documentation #47839

closed

BlueStore migration per-osd-device-copy section is misleading and incomplete

Added by Anthony D'Atri over 3 years ago. Updated over 2 years ago.

Status:
Resolved
Priority:
Normal
Assignee:
-
Category:
documentation
Target version:
-
% Done:

0%

Tags:
documentation
Backport:
Reviewed:
Affected Versions:
Pull request ID:

Description

https://www.spinics.net/lists/ceph-users/msg62734.html points out shortcomings in this section of the docs:

https://docs.ceph.com/en/latest/rados/operations/bluestore-migration/#per-osd-device-copy

It follows a lengthy section on repaving entire nodes at a time. The intent is to describe a process for migrating one OSD at a time without incurring backfill / network traffic outside of the given node. This is valuable, but as written it needs work.

Notably it refers to a copy function of ceph_objectstore_tool, which is not found in the sources. I suspect that dup is what's intended.

Moreover

Tooling not implemented!

Not documented!

isn't great. The process using dup should be tested and more fully documented.


Files

ceph-migrate-bluestore (9.42 KB) ceph-migrate-bluestore Chris Dunlop, 10/29/2020 11:50 PM
ceph-migrate-bluestore (9.23 KB) ceph-migrate-bluestore Chris Dunlop, 10/30/2020 01:29 AM
Actions #1

Updated by Anthony D'Atri over 3 years ago

That said, feel free to assign it to me unless someone else is ready to run with it. @zdover23?

We should along the way make it clear that an undeployed drive local to each node is required for this approach. I've long wanted to reserve a drive bay per system for just this sort of thing, but I suspect that capacity pressure means that very few people will have this luxury. One could undeploy an existing OSD and let it recover elsewhere; this sort of defeats the whole idea, but at least it's only one OSD recovering across the net vs. all of them.

Since 12 bay chassis are common and 13 bay chassis aren't, change the example to read 11/12 vs 12/13 to reflect this and be more explicit about the need to reserve or evacuate a drive bay.

Actions #2

Updated by Chris Dunlop over 3 years ago

An example of how to use ceph-objectstore-tool dup to convert from FileStore to BlueStore using a device copy: https://github.com/ceph/ceph/blob/master/qa/standalone/osd/osd-dup.sh

See attached ceph-migrate-bluestore which uses ceph-volume lvm prepare to get the basic structure for the new BlueStore version of the OSD.

"It Works For Me!"™

Actions #3

Updated by Chris Dunlop over 3 years ago

Sigh. You need to delete the old LV tags when adding the new tags, otherwise you end up with multiple tags, confusing ceph-volume lvm trigger. It somehow managed to work anyway for my first two OSDs but failed on the third.

Fixed ceph-migrate-bluestore attached.

Actions #4

Updated by Anthony D'Atri over 2 years ago

  • Status changed from New to Resolved
Actions

Also available in: Atom PDF